Results

Table 2 shows the SMR as a function of gender, age at stroke, diagnostic group, and duration of hospitalisation. Suicides were relatively more common among men (0.42%) than among women (0.20%), this difference being statistically significant (χ²(1)=43.9, p<0.001). However, as can be seen, this discrepancy conforms to the population base rate and there is no difference between the SMR values for men and women, both values being approximately 1.8 overall.

Almost 60% (214) of suicides occurred when the patient was over 65 years old. However, for age at stroke there is an almost uniform decrease in the SMR with increasing age. For patients suffering a stroke before the age of 50 the SMR indicate that there is almost a threefold excess of suicides relative to the base population. For patients suffering a stroke at age 80 or over the SMR is not significantly above unity.

Among the diagnostic groups, the intracerebral haemorrhage patients have a somewhat higher SMR than the other three groups, which do not differ among themselves. There is, however, substantial overlap in the confidence intervals for all diagnostic groups.

Duration of hospitalisation seems to be negatively associated with suicide risk. The SMR being highest for those hospitalised for less than one month and in fact falling below unity for those hospitalised for more than three months. It should be noted that there is only a weak correlation between age at stroke and duration of hospitalisation (Spearman's ρ = 0.14) and the effects of age and hospitalisation duration observed here are probably independent of each other. The additivity of the two effects is shown among 3428 patients who suffered a stroke before age 50 and who were hospitalised for less than two weeks. Within this group there were 17 suicides, yielding a SMR of 3.40 (95%CI 2.11, 5.47).

To examine the role of time since stroke we performed a Cox proportional hazards analysis.11 In this analysis all 114 098 patients were included, and cases were either observed until suicide (n=359) or were censored at death from other causes (n=65 724) or as being alive at the end of the study period—1 January 1994 (n=48 015). Gender, age at stroke (grouped as shown in table 2), diagnosis, and duration of hospitalisation (grouped as shown in table2) were all included as predictors. The computed hazard ratios are shown in table 3.

The highly significant hazard ratio for gender (p<0.001) confirms the approximately doubled greater risk of suicide among men than among women. The effect of age was also significant (p=0.04). Against a baseline of the group of patients of age 80 or older, the hazard ratios for all younger groups were significantly above unity with the exception of those aged between 60 and 69. There was no significant effect of diagnosis group (p>0.1). Within the groupings of duration of hospitalisation, patients hospitalised for more than three months had significantly lower hazard ratios than all three other groups. Figure 1 shows the overall cumulative proportional hazard over the maximum of 15 years covered by our study. For purposes of illustration the best fitting linear regression line has been superimposed. It can be seen that the function deviates from linearity. The negative acceleration indicates a relative concentration of suicides at under six years after a stroke.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Cumulative proportional hazard for suicide. A linear regression line is superimposed on the function.

Discussion

Although this study is both large scale and nationally representative there are a number of limitations that must be considered in the interpretation of the results. Firstly, although the diagnosis of stroke as recorded in the national hospitalisation register can be considered to be reliable, over 50% of the diagnoses were of undefined subtype. The age and sex distribution of these cases makes it probable that the large majority were ischaemic rather than haemorrhage, but at many smaller hospitals during the time period of the study neuroradiological investigations were not routinely performed, particularly for older patients. A recent study has, however, pointed to an apparent absence of relation between type of stroke and post-stroke depression.12 A related limitation of our register data is the lack of data on laterality of stroke. There have been a number of reports that post-stroke depression is more common in cases of left hemisphere stroke than right hemisphere.13 14 However, a recent comprehensive review of all relevant studies found no support for this assertion.15

It is often pointed out that the recording of suicide as a cause of death may be unreliable and under-reported. This is, in part, because it may in some circumstances be difficult to ascertain whether a death was a suicide or an accident, and given any uncertainty there may be a further unwillingness on the part of the relevant medical authority to conclude that death was caused by suicide, especially where there is strong religious disapprobation of suicide.16 17 There is, however, some evidence to suggest that these effects upon suicide statistics are, in fact, modest.18 Certainly it seems likely, however, that such factors would operate less in Denmark than in many other countries. Furthermore, there is no reason to expect that any bias that may be operating would differentially effect the recording of suicides among people who had earlier suffered a stroke, relative to the recording of suicides in the general population. Thus there would be no systematic bias in the mortality ratios as we are reporting them here.

In this study suicides occurred more often among men than among women and were more common among older rather than younger patients. Both of these characteristics are, however, in keeping with international epidemiological evidence for general populations18including Denmark.9 Population suicide rates in Denmark are close to figures reported for Western Europe and somewhat higher than those reported for North America.19

Our major specific finding is that the SMR for post-stroke patients are clearly increased relative to the general Danish population. This generally confirms the earlier findings of Stenageret al 6 for a single county in Denmark. It should however, be noted that the earlier study reported ratios much higher than were found here, particularly for ages below 60. For women under 60 years of age, a ratio of almost 14 was found, and for men under 60 the ratio was about 6. By contrast, we have found no difference in ratios between men and women, and the ratios for ages below 60 were under 3. It is difficult to attribute this anomaly to anything other than sampling fluctuation.

The overall trend in both the present and earlier studies was, however, similar, with ratios being higher for younger ages. This suggests that the disabilities consequent upon a stroke are more disturbing for younger patients than for older ones, as they would entail a greater change in life style with, for instance, the loss of ability to work. This is also consistent with the greater risk of post-stroke depression at younger ages.3

The effect of duration of hospitalisation is somewhat paradoxical, in that the risk of suicide decreases with duration of hospitalisation, indicating that increased stroke morbidity reduced the suicide risk. This may come about because the increased handicaps associated with more severe stroke reduce the capacity to carry out so decisive an action as suicide. Such an interpretation would be consistent with a reported positive association between level of functioning and suicide risk in spinal cord injury patients.20

The present evidence points to suicide risk being greatest within about the first five years after a stroke. This finding contrasts with that from a parallel study we have conducted on suicide after traumatic brain injury21 where suicide risk seemed to be relatively constant across the same 15 year period. The difference may have come about because suicides among traumatic brain injury patients are to some extent related to risk factors for the injury, for example, alcohol misuse, rather than the injury itself. In such cases suicide might be expected to be chronologically independent of the injury. In the present case there is little reason to expect that risk factors for stroke may also be risk factors for suicide, and the suicide is more likely to be reactive to the stroke and its consequences. It would seem, in keeping with findings concerning post-stroke depression,3 22 that the risk decreases with increasing years after stroke.

Although we have here emphasised the role of psychosocial consequences of stroke as causally implicated in the increased risks for suicide this should not be taken to discount possible neurochemical factors. A body of research has implicated the role of serotonergic dysfunction and also abnormalities of cortisol secretion in post-stroke depression.3

In summary it should be underlined that although the numbers of suicides in this study is comparatively low, the clearly increased rates relative to the age and gender matched population serve to underline the clinical need, as has been pointed out earlier,6 to be sensitive to the potential development of severe adverse emotional reactions to stroke. On present evidence, this is so particularly in relatively young patients, particularly in patients hospitalised for relatively shorter periods, and particularly in the first few years after the illness.